The development of advanced cyber threats and the geometric increase in security warnings have overwhelmed Security Operations Centers (SOCs), leading to alert fatigue, decision paralysis, and delayed threat response. Traditional rule-based systems lack contextual awareness and cannot scale as threats evolve, while deep learning models—though accurate—operate as black boxes unsuitable for high-stakes security environ-ments where decision transparency is critical for compliance and liability. This paper presents ThreatScope, a comprehensive hybrid threat intelligence platform that fuses the semantic embeddings of Sentence-BERT (SBERT) with TF-IDF statistical models through an intelligent voting ensemble, achieving 88% classification accuracy with 86% precision and 87% recall. The platform’s Dynamic Explainability Module (XAI) produces analyst-readable reasoning chains, attaining a 4.6/5 analyst trust rating—markedly superior to single-model baselines (2.9/5 SBERT, 2.3/5 TF-IDF). ThreatScope further implements enterprise-grade Role-Based Access Control (RBAC) with a three-tier permission hierarchy, immutable forensic logging via tamper-evident audit trails, and automated CVSS-based severity scoring. Deployed as a MERN-stack microservices architecture with a Python Flask AI backend and real-time Socket.IO collaboration, the system demonstrates a 57% reduction in Mean Time to Threat Response (MTTR: 210→90 minutes), 250% improvement in analyst throughput (6→18 threats/analyst/day), and 60% decrease in per-threat analysis cost. Comprehensive benchmarking against baseline models (TF-IDF+Na¨ıve Bayes, SVM, fine-tuned BERT) and commercial solutions (Jira, Splunk, Microsoft Sentinel) validates ThreatScope’s superior automated triage performance while preserving transparency and organizational compliance.